Authors: Gabriel Bellante*, Geospatial Technology and Applications Center
Topics: Natural Resources, Remote Sensing, Physical Geography
Keywords: vegetation mapping, forest structure, machine learning, optical remote sensing, Alaska
Session Type: Paper
Presentation File: No File Uploaded
A vegetation map of Alaska’s Kenai Peninsula was created in a multi-agency collaborative effort and produced with preexisting classification standards, providing baseline information to support project planning and management across the Kenai Peninsula. The final map is comprised of four distinct feature layers: 1) dominance type; 2) tree canopy cover; 3) tree size; and 4) tall shrub canopy cover. The dominance type map consists of 33 classes, including 28 vegetation classes and 5 classes of non-vegetation land cover types. Continuous canopy cover products were developed for areas classified as forest and tall shrub and a thematic layer depicting tree diameter class categories was generated for the forest class. Geospatial data, including remotely sensed imagery, topographic data, and gridded climate information, were assembled to; 1) produce a semi-automated image segmentation process to develop modeling units (mapping polygons), which represented relatively homogeneous areas of land cover and; 2) implement machine learning algorithms to develop vegetation models coupled with in situ reference data. Vegetation dominance type was best predicted using Sentinel 2 and Landsat 8 satellite imagery, while vegetation structure models were best informed using LiDAR and IfSAR elevation models. The map accuracy was assessed at the map group level using 400 photo interpreted calls and produced an overall accuracy of 80%. Accuracies will next be calculated at the dominance type level using in situ field observations from field crews and aerial helicopter surveys across Alaska’s Kenai Peninsula.